Acoustic emission data based deep learning approach for classification and detection of damage-sources in a composite panel

Structural health monitoring for lightweight complex composite structures is being investigated in this paper with a data-driven deep learning approach to facilitate automated learning of the map of transformed signal features to damage classes. Towards this, a series of acoustic emission (AE) based laboratory experiments have been carried out on a composite sample using a piezoelectric AE sensor network. The registered time-domain AE signals from the assigned sensor networks on the composite panel are processed with the continuous wavelet transform to extract time-frequency scalograms. A convolutional neural network based deep learning architecture is proposed to automatically extract the discrete damage features from the scalogram images and use them to classify damage-source regions in the composite panel. The proposed deep-learning approach has shown an effective damage monitoring potential with high training, validation and test accuracy for unseen datasets as well as for entirely new neighboring damage datasets. Further, the proposed network is trained, validated and tested only for the peak-signal data extracted from the raw AE data. The application of peak-signal scalogram data has shown a significant improvement in damage-source classification performance with high training, validation and test accuracy.